Research On The Image Super-resolution Based On Group Sparsity Dictionary

Image super-resolution reconstruction aims to estimate a high-resolution image fromone single image or a series of images of the same low-resolution scene. It can improvethe image quality and the space resolution of image in the premise without changing thehardware device. High resolution means that the image has the higher pixel density andcan provide more details.This paper investigates the principles and framework of the image super-resolutionreconstruction based on sparse representation. Considering of the advantages of thestructured dictionary, the mathematics model of the dictionary learning based on the groupsparsity will be constructed. An efficient algorithm for learning structured dictionaryaccording to the convex analysis and monotone operator theory will be proposed. Maincontents are as follows:Firstly, taking into account the importance of data dictionary of the above model, andthat the structured dictionary can improve the accuracy of the data representation, thepriori of the group sparsity will be extended to the mathematical model of the structureddictionary learning. Group sparsity means grouping the sparse representation vector,considering the sparsity of each group respectively to reflect the partial dependencebetween adjacent elements of the sparse representation vector. The partial dependence willbe reflected in the dictionary through the processing of the updating of the dictionary.Secondly, the existing forward-backward algorithm will be improved according to themonotone operator splitting method of the theory of convex optimization; An effectivealgorithm will be proposed for the above mathematical model of the structured dictionarylearning based on the group sparsity. The structured dictionary will be applied to the imagesuper-resolution reconstruction successfully.Finally, taking into account the real-tine requirements of the structured dictionary inpractical applications, an algorithm for the structured dictionary online updating based ongroup sparsity will be proposed through the derivation of the least square solution of thestructured dictionary. The updating of the dictionary dose not require iteration, just plus a matrix. The online updating process of the structured dictionary will be applied to theabnormality detection of the video successfully.